Phonon elastic wave damage detection method and device for ground wire detection

Abstract

The invention relates to the technical field of phonon elastic wave loss detection, and provides a phonon elastic wave loss detection method and device for ground wire detection, and the method comprises the following steps: employing a magnetostrictive sensor to collect an original phonon elastic wave signal of a ground wire, and carrying out the preprocessing of the original phonon elastic wave signal, and obtaining a phonon elastic wave signal; establishing an environmental factor influence model, collecting environmental parameters in real time, calculating a compensation coefficient through the environmental factor influence model, and compensating the phonon elastic wave signal based on the compensation coefficient to obtain a compensated phonon elastic wave signal; performing feature decomposition and dimension reduction processing on the compensation phonon elastic wave signal by adopting a distributed feature extraction model to obtain a defect identification feature vector, and inputting the defect identification feature vector into a defect identification model for defect identification to obtain a defect identification result of the ground wire; and outputting a defect detection report according to the defect identification result. Accurate recognition and efficient feature extraction of the ground wire defects are achieved, and the accuracy and efficiency of ground wire detection are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of phonon elastic wave detection for damage, and in particular, to a method and device for detecting phonon elastic waves for overhead ground wires. Background Art

[0002] As an important part of power transmission lines, the safe and stable operation of overhead ground wires is directly related to the reliability and safety of the power system. During long-term operation, various defects are likely to occur in overhead ground wires, such as broken strands, corrosion, wear, etc. If these defects are not discovered and processed in a timely manner, they may lead to a decline in the performance of overhead ground wires or even cause safety accidents. Usually, ultrasonic detection methods are used to detect overhead ground wires. Although ultrasonic detection methods can obtain the state information of overhead ground wires, there are many deficiencies in practical applications: the influence of environmental factors such as temperature, humidity, and electromagnetic interference is not fully considered during the detection process, resulting in the detection results being easily interfered by external conditions; the traditional signal processing methods have limited ability to suppress noise and lack intelligent feature extraction and analysis capabilities; the interpretation of detection results often relies on manual experience, making it difficult to achieve precise positioning and classification of defects.

[0003] The disadvantages of existing overhead ground wire detection technologies lie in the lack of consideration of environmental factors, resulting in low detection accuracy, inability to meet the actual needs of overhead ground wire condition monitoring, and severely restricting the efficiency of overhead ground wire detection. Therefore, there is an urgent need for a method for detecting phonon elastic waves for overhead ground wires that can overcome the above defects. Summary of the Invention

[0004] In view of this, the present invention proposes a method and device for detecting phonon elastic waves for overhead ground wires, which solves the problems of the existing technology that the lack of consideration of environmental factors leads to low detection accuracy, inability to meet the actual needs of overhead ground wire condition monitoring, and severely restricts the efficiency of overhead ground wire detection.

[0005] The technical solution of the present invention is realized as follows: In the first aspect, the present invention provides a method for detecting phonon elastic waves for overhead ground wires, including the following steps:

[0006] Collect the original phonon elastic wave signal of the overhead ground wire by using a magnetostrictive sensor, and preprocess the original phonon elastic wave signal to obtain a phonon elastic wave signal;

[0007] Establish an initial defect recognition model based on a deep learning network, and train and verify the initial defect recognition model through a historical phonon elastic wave data set to obtain a defect recognition model;

[0008] Establish an environmental factor influence model, collect environmental parameters in real time, calculate a compensation coefficient through the environmental factor influence model, and compensate the phonon elastic wave signal based on the compensation coefficient to obtain a compensated phonon elastic wave signal;

[0009] The compensated phonon elastic wave signal is subjected to feature decomposition and dimensionality reduction processing by using a distributed feature extraction model to obtain a defect recognition feature vector, and the defect recognition feature vector is input into a defect recognition model for defect recognition to obtain a defect recognition result of the ground wire;

[0010] A defect detection report is output according to the defect recognition result, and the defect detection report includes a defect type, a defect position, and a defect degree.

[0011] On the basis of the above technical solution, preferably, a magnetostrictive sensor is used to collect the original phonon elastic wave signal of the ground wire, and the original phonon elastic wave signal is preprocessed to obtain a phonon elastic wave signal, which specifically includes:

[0012] The original phonon elastic wave signal is subjected to multi-scale decomposition by using wavelet transform to obtain high-frequency wavelet coefficients, and the high-frequency wavelet coefficients are denoised by using an adaptive threshold method, and the denoised phonon elastic wave signal is reconstructed;

[0013] The denoised phonon elastic wave signal is segmented according to a fixed time window to obtain a segmented phonon elastic wave signal, the time-domain feature and the frequency-domain feature of each segmented phonon elastic wave signal are calculated, and wavelet packet decomposition is performed to obtain a phonon elastic wave signal.

[0014] On the basis of the above technical solution, preferably, an initial defect recognition model is established based on a deep learning network, and the initial defect recognition model is trained and verified by using a historical phonon elastic wave data set to obtain a defect recognition model, which specifically includes:

[0015] An initial defect recognition model is constructed based on a deep learning network model. The initial defect recognition model includes an input layer, a convolutional layer, a pooling layer, and a fully connected layer. Phonon elastic wave feature data with labeled tag information is collected from a historical phonon elastic wave database to obtain a historical phonon elastic wave data set. The historical phonon elastic wave data set is divided into a training set and a validation set. The input training set is normalized, the local features of the training set are extracted by the convolutional layer, the local features are dimensionally reduced and compressed by the pooling layer to obtain the key features of the training set, a dual attention mechanism is introduced, the weight of the key features is enhanced by the dual attention mechanism, the dual attention mechanism includes a channel attention mechanism and a spatial attention mechanism, and optimization training is performed according to a loss function, and the learning rate is dynamically adjusted to obtain a trained initial defect recognition model;

[0016] Calculate the confusion matrix of the initial defect recognition model on the validation set, evaluate the generalization ability of the initial defect recognition model, where the generalization ability includes the recognition accuracy of different types of defects, calculate the comprehensive evaluation index of the initial defect recognition model, and obtain the defect recognition model when the comprehensive evaluation index meets the preset threshold.

[0017] Based on the above technical solution, preferably, the initial defect recognition model is established based on a deep learning network, and the initial defect recognition model is trained and verified through a historical phonon elastic wave data set to obtain the defect recognition model, specifically including:

[0018] The calculation formula of the loss function is:

[0019] L total = L CE + α3L Focal + β3L Reg ;

[0020]

[0021] Among them, L total is the total loss, L CE is the cross-entropy loss, L Focal is the focal loss, L Reg is the regularization term, α3 and β3 are the weight coefficients of the cross-entropy loss and the regularization term respectively, N2 is the total number of samples, y jk is the label value of the j-th sample belonging to the k-th class, C is the total number of classes, p jk is the probability that the j-th sample belongs to the k-th class, η3 is the cross-entropy signal-to-noise ratio influence coefficient, μ3 is the cross-entropy signal-to-noise ratio attenuation coefficient, SNR3 is the cross-entropy signal-to-noise ratio, γ(SNR4) is the focal loss modulation parameter, SNR4 is the focal loss signal-to-noise ratio, γ0 is the focal loss basic modulation factor, γ1 is the focal loss signal-to-noise ratio adjustment coefficient, SNR0 is the reference signal-to-noise ratio, λ2 is the L1 regularization coefficient, λ3 is the L2 regularization coefficient, λ4 is the orthogonal constraint coefficient, ω0 is the regularization weight value, W is the set of model weights, W m is the weight matrix of the m-th layer, M is the total number of network layers, I is the identity matrix, is the norm;

[0022] The calculation formula of the comprehensive evaluation index is:

[0023]

[0024] Among them, Score is the comprehensive evaluation index, P is the accuracy rate, R is the recall rate, F1 is the F1 score, FP is the number of samples predicted as positive by the model but actually negative, N5 is the total number of samples, ω1, ω2, ω3, ω4 are the weight coefficients of the accuracy rate, recall rate, F1 score, and false alarm rate respectively, TP is the number of samples predicted as positive by the model and actually positive, and FN is the number of missed detection samples.

[0025] On the basis of the above technical solutions, preferably, the establishment of the environmental factor influence model, real-time collection of environmental parameters and calculation of the compensation coefficient through the environmental factor influence model, and compensation of the phonon elastic wave signal based on the compensation coefficient to obtain the compensated phonon elastic wave signal, specifically includes:

[0026] Using a temperature sensor, a humidity sensor, and an electromagnetic field intensity sensor to collect environmental parameters in real time, establishing a mapping relationship between environmental parameters and the attenuation characteristics of phonon elastic wave signals, obtaining the weight parameters of the environmental factor influence model through deep neural network training, and using the cross-validation method to verify and evaluate the environmental factor influence model;

[0027] The calculation formula of the environmental factor influence model is:

[0028] α env =w T ·f T (T)+w H ·f H (H)+w E ·f E (E);

[0029]

[0030]

[0031] Among them, α env is the environmental factor influence coefficient, w T 、w H 、w E are the environmental influence weights of temperature, humidity, and electromagnetic field respectively, f T (·)、f H (·)、f E (·)are the temperature influence function, humidity influence function, and electromagnetic field influence function respectively, k T 、k H 、k E are the temperature basic influence coefficient, humidity basic influence coefficient, and electromagnetic field intensity basic influence coefficient respectively, β T 、γ H 、λ EThey are the temperature adjustment coefficient, humidity adjustment coefficient, and electromagnetic field adjustment coefficient respectively. T, H, and E are the temperature value, humidity value, and electromagnetic field strength respectively. T0 and H0 are the temperature reference value and humidity reference value respectively;

[0032] Standardize the environmental parameters collected in real time, obtain the influence weights of each environmental parameter through the environmental factor influence model, calculate the compensation coefficient, and compensate the phonon elastic wave signal based on the compensation coefficient;

[0033] The calculation formula for compensating the phonon elastic wave signal is:

[0034]

[0035] φ comp =θ0·α env ·(1 + δ2·SNR5 -1 );

[0036] Where, S comp (t) is the compensated phonon elastic wave signal at time t, S orig (t) is the phonon elastic wave signal before compensation at time t, α env is the environmental factor influence coefficient, φ comp is the phase compensation angle, j is the imaginary unit, θ0 is the phase compensation reference value, δ2 is the phase signal-to-noise ratio adjustment coefficient, and SNR5 is the phase signal-to-noise ratio.

[0037] Based on the above technical solutions, preferably, the distributed feature extraction model is used to perform feature decomposition and dimensionality reduction processing on the compensated phonon elastic wave signal to obtain the defect recognition feature vector, and the defect recognition feature vector is input into the defect recognition model for defect recognition to obtain the defect recognition result of the ground wire, specifically including:

[0038] Perform multi-layer decomposition on the compensated phonon elastic wave signal using wavelet packet transform to obtain the compensated wavelet packet coefficients, use empirical mode decomposition to extract the intrinsic mode functions of the compensated phonon elastic wave signal, calculate the instantaneous frequency and energy entropy of each mode to obtain the intrinsic mode characteristics, and combine the compensated wavelet packet coefficients and intrinsic mode characteristics into a multi-dimensional feature matrix;

[0039] Perform preliminary dimensionality reduction on the multi-dimensional feature matrix using the principal component analysis method, retain the main feature components of the multi-dimensional feature matrix, use the t-SNE algorithm for further dimensionality reduction, maintain the local structure and global distribution of the main feature components, optimize the feature subset through the feature selection algorithm, generate the defect recognition feature vector, and input the defect recognition feature vector into the defect recognition model to obtain the defect recognition result of the ground wire.

[0040] Based on the above technical solutions, preferably, a defect detection report is output according to the defect recognition result, and the defect detection report includes defect type, defect location, and defect degree, specifically including:

[0041] Evaluate and analyze the defect recognition result, establish a defect feature database, quantitatively evaluate the defect type, defect location, and defect degree according to a preset evaluation standard, and obtain an evaluation result;

[0042] Generate a defect detection report based on the evaluation result, display the defect distribution in a visual manner through the defect detection report, and provide a defect risk level and maintenance suggestions.

[0043] In a second aspect, the present invention also provides a phonon elastic wave detection device for overhead ground wires, and the device includes:

[0044] A phonon elastic wave signal processing module, configured to collect the original phonon elastic wave signal of the overhead ground wire by using a magnetostrictive sensor, and preprocess the original phonon elastic wave signal to obtain a phonon elastic wave signal;

[0045] A defect recognition model module, configured to establish an initial defect recognition model based on a deep learning network, and train and verify the initial defect recognition model through a historical phonon elastic wave data set to obtain a defect recognition model;

[0046] An environmental factor compensation module, configured to establish an environmental factor influence model, collect environmental parameters in real time, calculate a compensation coefficient through the environmental factor influence model, and compensate the phonon elastic wave signal based on the compensation coefficient to obtain a compensated phonon elastic wave signal;

[0047] A defect recognition result module, configured to perform feature decomposition and dimensionality reduction processing on the compensated phonon elastic wave signal by using a distributed feature extraction model to obtain a defect recognition feature vector, and input the defect recognition feature vector into the defect recognition model for defect recognition to obtain the defect recognition result of the overhead ground wire;

[0048] A defect detection report module, configured to output a defect detection report according to the defect recognition result, and the defect detection report includes defect type, defect location, and defect degree.

[0049] In a third aspect, the present invention also provides an electronic device, including: at least one processor, at least one memory, a communication interface, and a bus;

[0050] Wherein, the processor, the memory, and the communication interface complete communication with each other through the bus, the memory stores program instructions executable by the processor, and the processor calls the program instructions to implement the steps of a phonon elastic wave detection method for overhead ground wires.

[0051] In a fourth aspect, the present invention also provides a computer-readable storage medium storing computer instructions that enable a computer to implement the steps of a method for detecting damage to a conductor and ground wire using phonon elastic waves.

[0052] The method and device for detecting damage to a conductor and ground wire using phonon elastic waves according to the present invention have the following beneficial effects compared with the prior art:

[0053] (1) By establishing an environmental factor influence model and adopting a signal preprocessing method for magnetostrictive sensors, the environmental interference problem is effectively solved and the signal quality is improved. Combining a deep learning network and a distributed feature extraction technology, accurate identification of conductor and ground wire defects and efficient feature extraction are achieved, and finally a complete detection report including defect type, defect location, and defect degree is output, improving the accuracy and efficiency of conductor and ground wire detection;

[0054] (2) By introducing a composite loss function including cross-entropy loss, focal loss, and a regularization term, and considering the influence factor of signal-to-noise ratio, precise optimization of the model training process is achieved. The focal loss enhances the attention to difficult-to-classify samples through modulation parameters, and the regularization term combines L1, L2 regularization, and orthogonal constraints to effectively prevent overfitting, thereby improving the recognition accuracy and generalization ability of the model;

[0055] (3) By establishing an environmental factor influence model including temperature, humidity, and electromagnetic field strength, and using a deep neural network to train and obtain model weight parameters, an accurate description of the mapping relationship between environmental parameters and the attenuation characteristics of phonon elastic wave signals is achieved, so that the compensation coefficient can be dynamically calculated according to real-time monitored environmental parameters, effectively eliminating the interference influence of environmental factors on phonon elastic wave signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0057] Figure 1 It is a flowchart of a method for detecting damage to a conductor and ground wire using phonon elastic waves according to the present invention;

[0058] Figure 2 It is a structural diagram of a device for detecting damage to a conductor and ground wire using phonon elastic waves according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0059] Next, in combination with the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0060] Please refer to Figure 1 , the present invention provides a method for detecting damage to a conductor and ground wire using phonon elastic waves, including the following steps:

[0061] Collect the original phonon elastic wave signal of the conductor and ground wire using a magnetostrictive sensor, and preprocess the original phonon elastic wave signal to obtain a phonon elastic wave signal. The preprocessing includes signal denoising, signal segmentation, and feature extraction;

[0062] Establish an initial defect recognition model based on a deep learning network, train and verify the initial defect recognition model through a historical phonon elastic wave data set to obtain a defect recognition model. The defect recognition model includes an input layer, a convolutional layer, a pooling layer, and a fully connected layer, and is provided with an attention mechanism module;

[0063] Establish an environmental factor influence model, collect environmental parameters in real time, and calculate a compensation coefficient through the environmental factor influence model. Compensate the phonon elastic wave signal based on the compensation coefficient to obtain a compensated phonon elastic wave signal. The environmental parameters include temperature, humidity, and electromagnetic interference;

[0064] Use a distributed feature extraction model to perform feature decomposition and dimensionality reduction processing on the compensated phonon elastic wave signal to obtain a defect recognition feature vector. Input the defect recognition feature vector into the defect recognition model for defect recognition to obtain a defect recognition result of the conductor and ground wire;

[0065] Output a defect detection report according to the defect recognition result. The defect detection report includes the defect type, defect location, and defect degree.

[0066] Specifically, in this embodiment, by establishing an environmental factor influence model and using a signal preprocessing method of a magnetostrictive sensor, the environmental interference problem is effectively solved and the signal quality is improved. Combining a deep learning network and a distributed feature extraction technology, accurate recognition of conductor and ground wire defects and efficient feature extraction are realized. Finally, a complete detection report including the defect type, defect location, and defect degree is output, improving the accuracy and efficiency of conductor and ground wire detection.

[0067] The step of collecting the original phonon elastic wave signal of the conductor and ground wire using a magnetostrictive sensor, and preprocessing the original phonon elastic wave signal to obtain a phonon elastic wave signal specifically includes:

[0068] Perform multi-scale decomposition on the original phonon elastic wave signal using wavelet transform, select appropriate wavelet basis functions and decomposition levels to obtain high-frequency wavelet coefficients, perform denoising on the high-frequency wavelet coefficients using the adaptive threshold method, and reconstruct to obtain the denoised phonon elastic wave signal;

[0069] The calculation formula of the adaptive threshold method is:

[0070]

[0071] where λ1 is the adaptive threshold, σ1 is the standard deviation of noise estimation, and N1 is the signal length;

[0072] Segment the denoised phonon elastic wave signal according to a fixed time window to obtain the segmented phonon elastic wave signal. Calculate the time-domain features (such as mean, variance, peak factor) and frequency-domain features (such as spectral energy distribution, main frequency component) for each segmented phonon elastic wave signal, and perform wavelet packet decomposition to obtain the phonon elastic wave signal;

[0073] The calculation formula of the peak factor of the time-domain feature is:

[0074]

[0075] where CF2 is the peak factor, x2(t2) is the signal sequence, SNR2 is the signal-to-noise ratio of the time-domain feature, N2 is the total number of signal sampling points, α2 is the first adjustment coefficient of the time-domain feature, and β2 is the second adjustment coefficient of the time-domain feature;

[0076] The calculation formula of the spectral energy of the frequency-domain feature is:

[0077]

[0078] where E i is the energy value of the i-th frequency band, X(f) is the Fourier transform result of f, W(·) is the window function, f is the frequency variable, f s is the sampling frequency, f c is the center frequency, and δ is the frequency attenuation coefficient.

[0079] Specifically, in this embodiment, the multi-scale decomposition of the original phonon elastic wave signal is performed by using wavelet transform, which can effectively separate different frequency components in the signal, thereby realizing the fine analysis of the signal. Appropriate wavelet basis functions and decomposition levels are selected to make the feature extraction of the signal more accurate. The adaptive threshold method is used for denoising processing, which can dynamically adjust the denoising intensity according to the actual situation of the signal, improve the signal-to-noise ratio of the signal, and reduce the interference of noise on the signal. The signal is segmented by a fixed time window, which is convenient for analyzing and extracting the local features of the signal. The time-domain features (such as mean, variance, peak factor) and frequency-domain features (such as spectral energy distribution, main frequency components) are calculated to achieve the comprehensive extraction of the signal features.

[0080] An initial defect recognition model is established based on a deep learning network. The initial defect recognition model is trained and verified through a historical phonon elastic wave data set to obtain a defect recognition model, which specifically includes:

[0081] An initial defect recognition model is constructed based on a deep learning network model. The initial defect recognition model includes an input layer, a convolutional layer, a pooling layer, and a fully connected layer. Phonon elastic wave feature data with labeled tag information is collected from a historical phonon elastic wave database to obtain a historical phonon elastic wave data set. The historical phonon elastic wave data set is divided into a training set and a validation set. The input training set is normalized. The local features of the training set are extracted through the convolutional layer. The local features are dimensionally reduced and compressed based on the pooling layer to obtain the key features of the training set. A dual attention mechanism is introduced, and the weight of the key features is enhanced through the dual attention mechanism. The dual attention mechanism includes a channel attention mechanism and a spatial attention mechanism. And it is optimized and trained according to the loss function, and the learning rate is dynamically adjusted to obtain the trained initial defect recognition model;

[0082] The confusion matrix of the initial defect recognition model on the validation set is calculated to evaluate the generalization ability of the initial defect recognition model. The generalization ability includes the recognition accuracy of different types of defects. The performance under different signal-to-noise ratio conditions is tested. The comprehensive evaluation index of the initial defect recognition model is calculated. When the comprehensive evaluation index meets the preset threshold, the defect recognition model is obtained.

[0083] Specifically, in this embodiment, an automated extraction and learning of phonon elastic wave signal features is achieved by constructing a deep learning network model with multiple layers; a dual attention mechanism (channel attention and spatial attention) is introduced to enhance the weights of key features and improve the model's recognition ability for different types of defect features; an optimized training method of dynamically adjusting the learning rate is adopted to improve the efficiency and stability of model training; the generalization ability of the model on the validation set is evaluated through a confusion matrix, and the performance under different signal-to-noise ratio conditions is considered to ensure that the model has good adaptability and anti-interference ability. Through the defect recognition model based on deep learning, not only the accuracy of defect recognition is improved, but also the model has strong generalization ability and environmental adaptability.

[0084] The initial defect recognition model is established based on the deep learning network, and the initial defect recognition model is trained and verified through the historical phonon elastic wave data set to obtain the defect recognition model, which specifically includes:

[0085] The calculation formula of the loss function is:

[0086] L total = L CE + α3L Focal + β3L Reg ;

[0087]

[0088] where, L total is the total loss, L CE is the cross-entropy loss, L Focal is the focal loss, L Reg is the regularization term, α3 and β3 are the weight coefficients of the cross-entropy loss and the regularization term respectively, N2 is the total number of samples, y jk is the label value of the j-th sample belonging to the k-th class, C is the total number of classes, p jk is the probability of the j-th sample belonging to the k-th class, η3 is the cross-entropy signal-to-noise ratio influence coefficient, μ3 is the cross-entropy signal-to-noise ratio attenuation coefficient, SNR3 is the cross-entropy signal-to-noise ratio, γ(SNR4) is the focal loss modulation parameter, SNR4 is the focal loss signal-to-noise ratio, γ0 is the focal loss basic modulation factor, γ1 is the focal loss signal-to-noise ratio adjustment coefficient, SNR0 is the reference signal-to-noise ratio, λ2 is the L1 regularization coefficient, λ3 is the L2 regularization coefficient, λ4 is the orthogonal constraint coefficient, ω0 is the regularization weight value, W is the set of model weights, W m is the weight matrix of the m-th layer, M is the total number of network layers, I is the identity matrix, is the norm;

[0089] The calculation formula of the comprehensive evaluation index is:

[0090]

[0091] Among them, Score is the comprehensive evaluation index, P is the accuracy rate, R is the recall rate, F1 is the F1 score, FP is the number of samples predicted as positive by the model but actually negative, N5 is the total number of samples, ω1, ω2, ω3, and ω4 are the weight coefficients of the accuracy rate, recall rate, F1 score, and false positive rate respectively, TP is the number of samples predicted as positive by the model and actually positive, and FN is the number of missed detection samples.

[0092] Specifically, in this embodiment, by introducing a composite loss function including cross-entropy loss, focal loss, and regularization term, the recognition difficulty of different types of defects is effectively balanced during the model training process. The cross-entropy loss is used to measure the difference between the model prediction and the actual label. The focal loss adjusts the attention to difficult-to-classify samples through modulation parameters, enhancing the model's recognition ability for minority-class samples. The regularization term combines L1 and L2 regularization to prevent the model from overfitting and improve the generalization ability of the model.

[0093] By calculating multiple evaluation parameters such as accuracy rate, recall rate, F1 score, and false positive rate, the performance of the model is comprehensively evaluated. The accuracy rate and recall rate measure the overall correctness of the model and the detection ability for defects respectively. The F1 score, as the harmonic mean of the two, provides a comprehensive evaluation of the model's performance. The false positive rate helps evaluate the reliability of the model in practical applications.

[0094] By setting a preset threshold for the comprehensive evaluation index, it is ensured that the performance of the model under different signal-to-noise ratio conditions reaches the expected standard. It not only improves the training effect of the model but also provides a reliable basis for performance evaluation, ensuring the stability of the model in practical applications.

[0095] The established environmental factor influence model collects environmental parameters in real time and calculates the compensation coefficient through the environmental factor influence model. Based on the compensation coefficient, the phonon elastic wave signal is compensated to obtain a compensated phonon elastic wave signal, which specifically includes:

[0096] Using a temperature sensor, a humidity sensor, and an electromagnetic field intensity sensor to collect environmental parameters in real time, establishing a mapping relationship between the environmental parameters and the attenuation characteristics of the phonon elastic wave signal, obtaining the weight parameters of the environmental factor influence model through deep neural network training, and using the cross-validation method to verify and evaluate the environmental factor influence model;

[0097] The calculation formula of the environmental factor influence model is:

[0098] α env =w T ·f T (T)+w H ·f H (H)+wE ·f E (E);

[0099]

[0100] Among them, α env is the environmental factor influence coefficient, w T , w H , w E are the environmental influence weights of temperature, humidity, and electromagnetic field respectively, f T (·), f H (·), f E (·) are the temperature influence function, humidity influence function, and electromagnetic field influence function respectively, k T , k H , k E are the temperature basic influence coefficient, humidity basic influence coefficient, and electromagnetic field strength basic influence coefficient respectively, β T , γ H , λ E are the temperature adjustment coefficient, humidity adjustment coefficient, and electromagnetic field adjustment coefficient respectively, T, H, and E are the temperature value, humidity value, and electromagnetic field strength respectively, and T0 and H0 are the temperature reference value and humidity reference value respectively;

[0101] Standardize the real-time collected environmental parameters, obtain the influence weights of each environmental parameter through the environmental factor influence model, comprehensively consider the coupling effect of multiple environmental factors, calculate the compensation coefficient, and compensate the phonon elastic wave signal based on the compensation coefficient;

[0102] The calculation formula for compensating the phonon elastic wave signal is:

[0103]

[0104] φ comp = θ0·α env ·(1 + δ2·SNR5 -1 );

[0105] Among them, S comp (t) is the compensated phonon elastic wave signal at time t, S orig (t) is the phonon elastic wave signal before compensation at time t, α env is the environmental factor influence coefficient, φ comp is the phase compensation angle, j is the imaginary unit, θ0 is the phase compensation reference value, δ2 is the phase signal-to-noise ratio adjustment coefficient, and SNR5 is the phase signal-to-noise ratio.

[0106] Specifically, in this embodiment, environmental parameters are collected in real time by using a temperature sensor, a humidity sensor, and an electromagnetic field intensity sensor to accurately capture the impact of environmental changes on the phonon elastic wave signal. A mapping relationship between environmental parameters and the attenuation characteristics of the phonon elastic wave signal is established, enabling the model to dynamically adjust the signal compensation strategy.

[0107] The weight parameters of the environmental factor influence model are trained using a deep neural network to ensure that the model can effectively learn and adapt to signal changes under different environmental conditions, improving the model's adaptability to complex environmental factors.

[0108] The environmental factor influence model is verified and evaluated through a cross-validation method to ensure the accuracy and stability of the model. The cross-validation method provides a reliable assessment of the model's generalization ability and reduces the risk of overfitting.

[0109] Compensating the phonon elastic wave signal based on the calculated compensation coefficient can effectively eliminate the interference of environmental factors on the signal and improve the authenticity and reliability of the signal.

[0110] The compensated phonon elastic wave signal is subjected to feature decomposition and dimensionality reduction processing using a distributed feature extraction model to obtain a defect recognition feature vector. The defect recognition feature vector is input into a defect recognition model for defect recognition to obtain the defect recognition result of the ground wire, specifically including:

[0111] The compensated phonon elastic wave signal is decomposed into multiple layers using wavelet packet transform to obtain compensated wavelet packet coefficients. The empirical mode decomposition is used to extract the intrinsic mode functions of the compensated phonon elastic wave signal, and the instantaneous frequency and energy entropy of each mode are calculated to obtain intrinsic mode features. The compensated wavelet packet coefficients and intrinsic mode features are combined into a multi-dimensional feature matrix;

[0112] The calculation formula for the multi-layer decomposition is:

[0113]

[0114] Where is the time-frequency feature function, and are the time variable and frequency variable respectively, WPT p,q (s) is the compensated wavelet packet coefficient of the qth frequency band in the pth layer, is the wavelet basis function, IMF r (·) is the rth intrinsic mode function, is the instantaneous frequency of the rth intrinsic mode function, and j is the imaginary unit;

[0115] The principal component analysis method is used to preliminarily reduce the dimension of the multi-dimensional feature matrix, retaining the main feature components of the multi-dimensional feature matrix. The t-SNE algorithm is used to further reduce the dimension, maintaining the local structure and global distribution of the main feature components. The feature subset is optimized through the feature selection algorithm to generate the defect recognition feature vector. The defect recognition feature vector is input into the defect recognition model to obtain the defect recognition result of the ground wire;

[0116] The feature matrix after dimensionality reduction is:

[0117] Y = t-SNE(PCA(X))·W;

[0118] Where Y is the feature matrix after dimensionality reduction, X is the multi-dimensional feature matrix, PCA(·) is the principal component analysis method, and t-SNE(·) is the t-distributed stochastic neighbor embedding algorithm.

[0119] Specifically, in this embodiment, the compensated phonon elastic wave signal is decomposed into multiple layers by using wavelet packet transform, which can effectively capture the characteristics of the signal at different frequencies and time scales. The wavelet packet transform provides higher frequency resolution, enabling the detailed characteristics of the signal to be retained and extracted. The empirical mode decomposition (EMD) is used to extract the intrinsic mode functions of the compensated phonon elastic wave signal, and the instantaneous frequency and energy entropy of each mode are calculated, which can reveal the internal dynamic characteristics of the signal, help identify the non-linear and non-stationary characteristics in the signal, and improve the accuracy of defect recognition.

[0120] Combining the compensated wavelet packet coefficients and the intrinsic mode characteristics into a multi-dimensional feature matrix provides rich feature information, can more comprehensively describe the signal characteristics, and improves the input quality of the defect recognition model.

[0121] The multi-dimensional feature matrix is dimensionally reduced by the principal component analysis method and the t-distributed stochastic neighbor embedding (t-SNE) algorithm, reducing the feature dimension, lowering the computational complexity, and at the same time retaining the important feature information.

[0122] The defect detection report is output according to the defect recognition result. The defect detection report includes the defect type, defect location, and defect degree, specifically including:

[0123] The defect recognition result is evaluated and analyzed, a defect feature database is established, and the defect type, defect location, and defect degree are quantitatively evaluated according to the preset evaluation criteria to obtain the evaluation result;

[0124] A multi-dimensional evaluation matrix including defect feature parameters, evaluation indicators, and risk levels is established. The fuzzy comprehensive evaluation method is used to calculate the defect severity index, the defect development trend is determined based on historical data statistics analysis, and the defect location is accurately positioned in combination with the structural characteristics of the ground wire.

[0125] Generate a defect detection report based on the evaluation results, and display the defect distribution in a visual manner through the defect detection report, and provide the defect risk level and maintenance suggestions.

[0126] Design the report template using a hierarchical structure, including three levels: overview, detailed analysis, and suggestions. Use heat maps and 3D models to display the defect distribution, give early warnings of defect development based on a deep learning prediction model, and automatically generate targeted maintenance plans and treatment suggestions.

[0127] In a specific embodiment, during the detection process of the ground wire, a defect recognition model is used to analyze the phonon elastic wave signal to obtain preliminary defect recognition results. These results include the identified defect type, defect location, and defect degree.

[0128] Store the identified defect information in a dedicated database, which contains historical defect data and corresponding feature information. The establishment of the database helps subsequent defect analysis and model optimization.

[0129] Quantitatively evaluate the identified defects according to preset evaluation criteria. The evaluation criteria may include the severity of the defects, the impact on the function of the ground wire, and the priority of repair.

[0130] Generate a detailed evaluation result report through the quantitative evaluation of the defects. The report details the type, location, degree, and recommended treatment measures of each defect.

[0131] Specifically, through the evaluation and analysis of the recognition results in this embodiment, the accuracy of the defect recognition model can be verified and improved. The evaluation process can identify the deficiencies of the model and provide data support for the further optimization of the model.

[0132] The establishment and use of the defect feature database enable historical data to be used for the analysis and decision support of current defects. Through quantitative evaluation, an objective standard is provided to judge the severity and priority of defects, thereby optimizing maintenance and repair strategies.

[0133] Through the detailed defect evaluation report, maintenance personnel can more accurately understand the health status of the ground wire. Timely defect identification and evaluation help prevent potential failures and ensure the reliable operation of the ground wire.

[0134] The evaluation results and historical data in the database can be used to train and improve the defect recognition model, making it have better adaptability and accuracy in different environments and conditions.

[0135] Please refer to Figure 2 , the present invention also provides a phonon elastic wave detection device for ground wire detection, and the device includes:

[0136] A phonon elastic wave signal processing module, which is used to collect the original phonon elastic wave signal of the ground wire by using a magnetostrictive sensor, and preprocess the original phonon elastic wave signal to obtain a phonon elastic wave signal;

[0137] A defect recognition model module, which is used to establish an initial defect recognition model based on a deep learning network, and train and verify the initial defect recognition model through a historical phonon elastic wave data set to obtain a defect recognition model;

[0138] An environmental factor compensation module, which is used to establish an environmental factor influence model, collect environmental parameters in real time, calculate a compensation coefficient through the environmental factor influence model, and compensate the phonon elastic wave signal based on the compensation coefficient to obtain a compensated phonon elastic wave signal;

[0139] A defect recognition result module, which is used to perform feature decomposition and dimensionality reduction processing on the compensated phonon elastic wave signal by using a distributed feature extraction model to obtain a defect recognition feature vector, and input the defect recognition feature vector into the defect recognition model for defect recognition to obtain a defect recognition result of the ground wire;

[0140] A defect detection report module, which is used to output a defect detection report according to the defect recognition result, and the defect detection report includes a defect type, a defect location, and a defect degree.

[0141] Specifically, the phonon elastic wave signal processing module of this embodiment uses a magnetostrictive sensor to collect the original phonon elastic wave signal of the ground wire and preprocesses it. Through efficient signal acquisition and preprocessing (including noise reduction, segmentation, and feature extraction), the quality and reliability of the signal are improved.

[0142] The defect recognition model establishes an initial defect recognition model based on a deep learning network, and trains and verifies it through a historical phonon elastic wave data set. By using the powerful feature learning ability of deep learning, the model can automatically extract and identify complex defect features, improving the accuracy and efficiency of defect recognition. The introduction of the attention mechanism further enhances the model's attention to key features and improves the recognition performance.

[0143] The environmental factor influence model module collects environmental parameters (such as temperature, humidity, and electromagnetic interference) in real time, and calculates a compensation coefficient through the environmental factor influence model. By monitoring and compensating environmental factors in real time, it can effectively eliminate the interference of the environment on the signal and ensure the stability and accuracy of the signal.

[0144] The feature extraction and dimensionality reduction module performs feature decomposition and dimensionality reduction processing on the compensated phonon elastic wave signal. Through methods such as wavelet packet transform and empirical mode decomposition, the system can extract rich signal features and reduce the computational complexity through dimensionality reduction technology, improving the processing speed and efficiency of the model.

[0145] The defect evaluation and report generation module evaluates and analyzes the defect recognition results and generates a defect detection report. By establishing a defect feature database and a multi-dimensional evaluation matrix, the system can comprehensively quantify and trend-analyze the defects, provide a detailed defect detection report and maintenance suggestions, enhancing the system's decision-making support ability.

[0146] The present invention also discloses an electronic device, including: at least one processor, at least one memory communication interface, and a bus: wherein, the processor, the memory, and the communication interface complete mutual communication through the bus; the memory stores program instructions executable by the processor, and the processor invokes the program instructions to implement a distributed photovoltaic power station unmanned aerial vehicle inspection method.

[0147] The present invention also discloses a computer-readable storage medium, which stores computer instructions that enable the computer to implement all or part of the steps of the distributed photovoltaic power station unmanned aerial vehicle inspection method described in the embodiments of the present invention. The storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory ROM, random access memory RAM, magnetic disks, or optical discs that can store program codes.

[0148] The above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A phonon elastic wave detection method for detecting ground wires and conductors, characterized in that The method includes the following steps: Collect the original phonon elastic wave signal of the ground wire by using a magnetostrictive sensor, and preprocess the original phonon elastic wave signal to obtain a phonon elastic wave signal; Establish an initial defect recognition model based on a deep learning network, and train and verify the initial defect recognition model through a historical phonon elastic wave data set to obtain a defect recognition model; Establish an environmental factor influence model, collect environmental parameters in real time, calculate a compensation coefficient through the environmental factor influence model, and compensate the phonon elastic wave signal based on the compensation coefficient to obtain a compensated phonon elastic wave signal; Use a distributed feature extraction model to perform feature decomposition and dimensionality reduction processing on the compensated phonon elastic wave signal to obtain a defect recognition feature vector, input the defect recognition feature vector into the defect recognition model for defect recognition, and obtain a defect recognition result of the ground wire; Output a defect detection report according to the defect recognition result, where the defect detection report includes a defect type, a defect position, and a defect degree.

2. The phonon elastic wave detection and loss detection method for ground wires according to claim 1, wherein The step of collecting the original phonon elastic wave signal of the ground wire by using a magnetostrictive sensor and preprocessing the original phonon elastic wave signal to obtain a phonon elastic wave signal specifically includes: Perform multi-scale decomposition on the original phonon elastic wave signal by using wavelet transform to obtain high-frequency wavelet coefficients, perform denoising on the high-frequency wavelet coefficients by using an adaptive threshold method, and reconstruct to obtain a denoised phonon elastic wave signal; Segment the denoised phonon elastic wave signal according to a fixed time window to obtain a segmented phonon elastic wave signal, calculate the time-domain feature and the frequency-domain feature for each segmented phonon elastic wave signal, and perform wavelet packet decomposition to obtain a phonon elastic wave signal.

3. The phonon elastic wave detection and damage detection method for ground wires and overhead lines according to claim 2, characterized in that, The step of establishing an initial defect recognition model based on a deep learning network and training and verifying the initial defect recognition model through a historical phonon elastic wave data set to obtain a defect recognition model specifically includes: Construct an initial defect recognition model based on a deep learning network model. The initial defect recognition model includes an input layer, a convolutional layer, a pooling layer, and a fully connected layer. Collect phonon elastic wave feature data with labeled tag information from a historical phonon elastic wave database to obtain a historical phonon elastic wave data set. Divide the historical phonon elastic wave data set into a training set and a verification set. Perform normalization processing on the input training set. Extract the local features of the training set through the convolutional layer. Perform dimensionality reduction and compression on the local features based on the pooling layer to obtain the key features of the training set. Introduce a dual attention mechanism to enhance the weight of the key features through the dual attention mechanism. The dual attention mechanism includes a channel attention mechanism and a spatial attention mechanism, and perform optimized training according to a loss function, dynamically adjust the learning rate, and obtain a trained initial defect recognition model; Calculate the confusion matrix of the initial defect recognition model on the verification set, evaluate the generalization ability of the initial defect recognition model. The generalization ability includes the recognition accuracy of different types of defects, calculate the comprehensive evaluation index of the initial defect recognition model, and obtain a defect recognition model when the comprehensive evaluation index meets a preset threshold.

4. The phonon elastic wave detection and loss detection method for ground wires according to claim 3, characterized in that, An initial defect recognition model is established based on a deep learning network, and the initial defect recognition model is trained and verified by a historical phonon elastic wave data set to obtain a defect recognition model, which specifically includes: The calculation formula of the loss function is: L total = L CE + α3L Focal + β3L Reg ; Among them, L total is the total loss, L CE is the cross-entropy loss, L Focal is the focal loss, L Reg is the regularization term, α3 and β3 are the weight coefficients of the cross-entropy loss and the regularization term respectively, N2 is the total number of samples, y jk is the label value of the j-th sample belonging to the k-th class, C is the total number of classes, p jk is the probability that the j-th sample belongs to the k-th class, η3 is the cross-entropy signal-to-noise ratio influence coefficient, μ3 is the cross-entropy signal-to-noise ratio attenuation coefficient, SNR3 is the cross-entropy signal-to-noise ratio, γ(SNR4) is the focal loss modulation parameter, SNR4 is the focal loss signal-to-noise ratio, γ0 is the focal loss base modulation factor, γ1 is the focal loss signal-to-noise ratio adjustment coefficient, SNR0 is the reference signal-to-noise ratio, λ2 is the L1 regularization coefficient, λ3 is the L2 regularization coefficient, λ4 is the orthogonal constraint coefficient, ω0 is the regularization weight value, W is the set of model weights, W m is the weight matrix of the m-th layer, M is the total number of network layers, I is the identity matrix, is the norm; The calculation formula of the comprehensive evaluation index is: Where Score is the comprehensive evaluation index, P is the accuracy rate, R is the recall rate, F1 is the F1 score, FP is the number of samples predicted as positive by the model but actually negative, N5 is the total number of samples, ω1, ω2, ω3, ω4 are the weight coefficients of the accuracy rate, recall rate, F1 score, and false alarm rate respectively, TP is the number of samples predicted as positive by the model and actually positive, and FN is the number of missed detection samples.

5. The phonon elastic wave detection and loss detection method for ground wires and overhead lines according to claim 1, characterized in that An environmental factor influence model is established, environmental parameters are collected in real time, and a compensation coefficient is calculated through the environmental factor influence model. The phonon elastic wave signal is compensated based on the compensation coefficient to obtain a compensated phonon elastic wave signal, which specifically includes: A temperature sensor, a humidity sensor, and an electromagnetic field intensity sensor are used to collect environmental parameters in real time, a mapping relationship between environmental parameters and the attenuation characteristics of the phonon elastic wave signal is established, the weight parameters of the environmental factor influence model are obtained through training by a deep neural network, and the environmental factor influence model is verified and evaluated by a cross-validation method; The calculation formula of the environmental factor influence model is: α env = w T · f T (T)+ w H · f H (H)+ w E · f E (E); Among them, α env is the environmental factor influence coefficient, w T , w H , w E are the environmental influence weights of temperature, humidity, and electromagnetic field respectively, f T (·), f H (·), f E (·) are the temperature influence function, humidity influence function, and electromagnetic field influence function respectively, k T , k H , k E are the temperature basic influence coefficient, humidity basic influence coefficient, and electromagnetic field strength basic influence coefficient respectively, β T , γ H , λ E are the temperature adjustment coefficient, humidity adjustment coefficient, and electromagnetic field adjustment coefficient respectively, T, H, and E are the temperature value, humidity value, and electromagnetic field strength respectively, and T0 and H0 are the temperature reference value and humidity reference value respectively; The environmental parameters collected in real time are standardized, the influence weights of each environmental parameter are obtained through the environmental factor influence model, the compensation coefficient is calculated, and the phonon elastic wave signal is compensated based on the compensation coefficient; The calculation formula of the compensated phonon elastic wave signal is: φ comp = θ0·α env ·(1 + δ2·SNR5 -1 ); Among them, S comp (t) is the compensated phonon elastic wave signal at time t, and S orig (t) is the phonon elastic wave signal before compensation at time t. α env is the environmental factor influence coefficient, φ comp is the phase compensation angle, j is the imaginary unit, θ0 is the phase compensation reference value, δ2 is the phase signal-to-noise ratio adjustment coefficient, and SNR5 is the phase signal-to-noise ratio.

6. The phonon elastic wave detection and damage detection method for ground wire and overhead line as claimed in claim 1, wherein The compensated phonon elastic wave signal is subjected to feature decomposition and dimensionality reduction processing by using a distributed feature extraction model to obtain a defect recognition feature vector, and the defect recognition feature vector is input into the defect recognition model for defect recognition to obtain the defect recognition result of the ground wire, which specifically includes: The compensated phonon elastic wave signal is decomposed into multiple layers by using wavelet packet transform to obtain compensated wavelet packet coefficients, the intrinsic mode functions of the compensated phonon elastic wave signal are extracted by using empirical mode decomposition, the instantaneous frequency and energy entropy of each mode are calculated to obtain the intrinsic mode features, and the compensated wavelet packet coefficients and the intrinsic mode features are combined into a multi-dimensional feature matrix; The multi-dimensional feature matrix is initially reduced in dimension by using the principal component analysis method to retain the main feature components of the multi-dimensional feature matrix, and the t-SNE algorithm is used for further dimensionality reduction to maintain the local structure and global distribution of the main feature components. The feature subset is optimized by a feature selection algorithm to generate a defect recognition feature vector, and the defect recognition feature vector is input into the defect recognition model to obtain the defect recognition result of the ground wire.

7. The phonon elastic wave detection and loss detection method for ground wires and overhead lines according to claim 1, characterized in that A defect detection report is output according to the defect recognition result, and the defect detection report includes the defect type, defect location, and defect degree, which specifically includes: The defect recognition result is evaluated and analyzed, a defect feature database is established, and the defect type, defect location, and defect degree are quantitatively evaluated according to a preset evaluation standard to obtain an evaluation result; Generate a defect detection report based on the evaluation results, and display the defect distribution in a visual manner through the defect detection report, and provide the defect risk level and maintenance suggestions.

8. A phonon elastic wave detecting and damage detecting device for guide wires and ground wires, characterized in that The device includes: A phonon elastic wave signal processing module, configured to collect the original phonon elastic wave signal of the ground wire by using a magnetostrictive sensor, and preprocess the original phonon elastic wave signal to obtain a phonon elastic wave signal; A defect recognition model module, configured to establish an initial defect recognition model based on a deep learning network, and train and verify the initial defect recognition model through a historical phonon elastic wave data set to obtain a defect recognition model; An environmental factor compensation module, configured to establish an environmental factor influence model, collect environmental parameters in real time, calculate a compensation coefficient through the environmental factor influence model, and compensate the phonon elastic wave signal based on the compensation coefficient to obtain a compensated phonon elastic wave signal; A defect recognition result module, configured to perform feature decomposition and dimensionality reduction processing on the compensated phonon elastic wave signal by using a distributed feature extraction model to obtain a defect recognition feature vector, and input the defect recognition feature vector into the defect recognition model for defect recognition to obtain a defect recognition result of the ground wire; A defect detection report module, configured to output a defect detection report according to the defect recognition result, where the defect detection report includes a defect type, a defect location, and a defect degree.

9. An electronic device, characterized in that, Includes: At least one processor, at least one memory, a communication interface, and a bus; Wherein, the processor, the memory, and the communication interface complete communication with each other through the bus, the memory stores program instructions executable by the processor, and the processor calls the program instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, and the computer instructions cause the computer to implement the method according to any one of claims 1 to 7.

Images